- Record metadata at end of run
- Read metadata at start of run
- Running in date groups, ordering by fastest
- record and compare run time on examples
- record run time on groups
- merge on metadata, don't overwrite
- sort the groups as well. our orderer will get called with groups and then with examples in the groups
- parent groups will reflect the most relevant of their deepest children
- refactoring: make RunResults a class instead of a hash. Should clean up a lot